DeepFilterNet vs RNNoise
DeepFilterNet
github.comRNNoise
gitlab.xiph.orgSide by side
| Figure | DeepFilterNet | RNNoise |
|---|---|---|
| Price from | Free | Free |
| Free plan | Yes | Yes |
| Free trial | — | — |
| Open source | Yes | — |
| Windows | Yes | — |
| macOS | Yes | — |
| Linux | Yes | Yes |
| API | Yes | — |
| Self-hosted | Yes | — |
| Plans published | 2 | 1 |
| Spec score · AI Noise Cancellation Software | 9.0 (#1) | 7.1 (#20) |
| Distribution | The project provides precompiled binaries through its GitHub releases and a Python wheel through PyPI. | The README says the GitHub repository is a convenience copy and identifies GitLab as the location of the latest source. |
| License | The repository code is free and open source under either the MIT License or Apache License 2.0, at the user's option. | The repository identifies its license as BSD-3-Clause. |
| Purpose | DeepFilterNet is a low-complexity speech enhancement framework for full-band 48 kHz audio using deep filtering. | RNNoise is a noise suppression library based on a recurrent neural network. |
| Security | The repository page shows a Security and quality tab but provides no security or compliance claims in the project information read. | The opened project pages state a software license but do not describe security certifications or compliance. |
| Training | Training expects datasets in HDF5 format, with a dataset configuration file for train, validation, and test sets. | Model training requires clean speech and noise data sampled at 48 kHz in 16-bit PCM format. |
Plans
In their own words
DeepFilterNet
DeepFilterNet is a free, open-source framework for speech enhancement on full-band 48 kHz audio, using deep filtering. Its command-line tool suppresses noise in WAV files, with 48 kHz as its only supported sampling rate. A LADSPA plugin provides real-time noise suppression and can be used with PipeWire filter chains to create a virtual microphone or audio output, including for VoIP apps such as Zoom or Discord. The plugin's minimum latency is 20 ms from STFT processing, plus any delay added by the LADSPA host. The Rust command-line tool defaults to DeepFilterNet2; the Python interface documents DeepFilterNet and DeepFilterNet2. An optional post-filter may slightly over-attenuate very noisy sections. The framework supports Linux, macOS and Windows, though training is tested only under Linux. The Python package installs as deepfilternet, and an optional train extra adds data-loading functionality for Linux training. Code is available under either the MIT License or Apache License 2.0.